🎓 Student Performance Classification

🚀 Machine Learning Project | Model Evaluation + Polynomial Features + SHAP Interpretation

🌟 Level-1 → Task 1 + Bonus Completed ✅

https://elevvomlstudentperformancefactors-polymonialregression.streamlit.app/

---

📌 Task Description

Build a model to predict students' exam scores based on their study hours Perform data cleaning and basic visualization to understand the dataset Split the dataset into training and testing sets Train a linear regression model to estimate final scores Visualize predictions and evaluate model performance

---

📂 Dataset

• Source: Student Performance Classification

• File Used: StudentPerformanceFactors.csv

• Target Variable: Score_Category (Categorized from Exam_Score into performance levels)

   Features Include:
   Attendance
   Hours_Studied
   Previous_Scores
   Sleep_Hours
   Physical_Activity
   Parental_Involvement
   Motivation_Level
   Family_Income
   Teacher_Quality
   Gender
   etc.
  

---

🧰 Tools & Libraries Used

• 🐍 Python
• 📊 Pandas & NumPy
• 📈 Matplotlib & Seaborn
• 📊 Scikit-learn
• 📊 SHAP (Model Explainability)

---

🔄 Project Workflow

✅ 1. Data Overview & Cleaning

• Checked missing values
• Verified data types
• Converted categorical features
• Created Score_Category as classification target

---

✅ 2. Exploratory Data Analysis (EDA)

📈 Visualization of the distribution

• Histogram
• Matrix Correlation
• Boxplot
• Scatter plot + Regression Line

📊 Statistical Testing

• ANOVA → for numerical vs target
• Chi-Square → for categorical vs target

📌 Insight: Attendance and Previous Scores showed strong statistical significance.

---

✅ 3. Feature Engineering

• Academic Effort Score
• Academic Effort Score
• Learning Momentum
• Academic Risk Index

✔ Combines related indicators into interpretable composite scores

---

✅ 4. Encoding & Scaling

• One-Hot Encoding for categorical features
• StandardScaler for numerical normalization

---

✅ 5. Model Testing (6 Algorithms Compared)

Models Evaluated:

Model Accuracy	Accuracy Mean	ROC-AUC Mean
Logistic Regression	95.02	98.99
SVM	91.64	98.22
Gradient Boosting	85.18	95.20
Random Forest	80.57	93.13
KNN	73.55	88.70
Decision Tree	72.85	78.98

📌 Best Baseline Model: Logistic Regression 💡

---

✅ 6. Feature Importance - Logistic Regression

Selected top k most significant features before modeling.

• Attendance
• Motivation_Level_Low
• Hours_Studied
• Previous_Scores

📌 The primary determinants of academic performance.

---

✅ 7. Hyperparameter Tuning

Used GridSearchCV on Logistic Regression.

🔍 Best Parameters:

C = 100
Penalty = L1
Solver = saga
Class_weight = balanced

Best CV Accuracy: 95.34%

📌 L1 regularization helped in automatic feature selection.

---

✅ 8. Polynomial Feature

• Adding Polynomial Features (degree=2)
• Training accuracy (99.02%)
• Test Accuracy (91.07%),
• (Indicating overfitting))

📌The linear Logistic Regression model without Polynomial transformation provides a better bias-variance balance for this dataset.

---

📊 Final Model Performance

• ROC-AUC (Multiclass OVR) evaluated
• Balanced class handling applied

---

✅ 9. Model Interpretation (SHAP Analysis)

Applied SHAP to understand feature contribution.

🔎 Key Findings:

• Attendance has the highest positive impact
• Academic Effort & Study Hours, Study habits coaching program
• Previous Scores, Remedial program based on previous performance

📌 SHAP confirms statistical and modeling results.

---

📊 Business Insight

• ✔ Improving attendance discipline
• ✔ Optimizing learning hours and quality
• ✔ Remedial programs based on previous grades
• ✔ Increasing access to learning resources
• ✔ L1 regularization enhances interpretability
• ✔ Strengthening parental support

In conclusion, interventions based on study habits and environmental support will have the greatest impact on improving student performance compared to demographic factors alone.

---

📚 Concepts Covered

• 📊 Statistical Hypothesis Testing
• 🔺 Polynomial Feature Engineering
• 📊 Feature Selection (ANOVA)
• 🤖 Classification Modeling
• ⚙ Hyperparameter Tuning
• 📈 Model Comparison
• 📉 Multiclass ROC-AUC
• 🧠 Explainable AI (SHAP)

---